FinOps for Private Cloud

The calculation rests on unit cost models that are built rather than observed: cost per GB stored, cost per vCPU, cost per transaction, cost per tenant. A single physical rack server illustrates the difficulty well: once virtualised, it can simultaneously host VMs dedicated to storage and others to compute, consumed by different teams. Without an explicit allocation model, the cost of that rack remains an indivisible block, impossible to attribute fairly to its actual consumers.

Establishing those unit economics requires allocating fixed costs across variable consumption, an exercise Public Cloud resolves natively through usage-based pricing, and that Private Cloud must rebuild through modelling. This allocation goes well beyond hardware depreciation alone: it must cover the full set of real operating costs of a data centre, physical and logical security, insurance, maintenance contracts, electricity, cooling and water consumption, transport and logistics of the hardware. Omitting any one of these components structurally understates the real cost of Private Cloud and biases any subsequent arbitrage with Public Cloud.

TCO per workload, covering infrastructure, software and labour across the full lifecycle, becomes the reference indicator to objectively compare an internally hosted workload with its Public Cloud equivalent.

Optimisation in Private Cloud is not about adjusting a contractual discount rate, it targets the physical utilisation of the owned asset. Physical-to-virtual consolidation ratio, identification of under-used or orphaned servers through audit and hardware monitoring, controlled decommissioning of obsolete equipment: every point of consolidation gained has a direct impact on TCO, with no external negotiation available.

Procurement follows a different logic too: moving from traditional bulk purchasing to just-in-time provisioning, reducing the lag between purchase request and operational availability, is one of the few levers directly actionable on total cost.

Other levers, specific to Private Cloud, remain often underexploited. Renegotiating hypervisor licensing contracts has become a critical item: recent market pricing shifts (notably VMware/Broadcom) have made it one of the first variable cost lines in a virtualised environment, on a par with a Public Cloud commitment. The arbitrage between extending a server's life beyond its accounting depreciation and replacing it must factor in the energy overhead of ageing hardware, often ignored in a purely accounting reading. Storage tiering, the split between high-performance capacity and cold storage based on the actual criticality of the data, is a workstream rarely audited once the initial architecture is in place.

Two complementary levers, at the intersection of Facilities and IT, are frequently overlooked: cooling optimisation (hot/cold aisle containment, free cooling), whose impact on PUE often exceeds that of compute optimisation itself, and the valorisation of decommissioned hardware through resale or refurbishment, which turns a pure cost centre into residual value recovery. Finally, the redundancy levels selected (N+1, N+2) are often frozen historically and never reassessed against the actual criticality of the workloads they protect, even though they weigh directly on the sizing, and therefore the cost, of the entire infrastructure.